Deep learning-based design of long-wave infrared dielectric grating accelerator structures

Dielectric laser accelerators (DLAs), as compact particle accelerators, rely critically on their structural design to determine both the energy gain and beam quality of accelerated bunches. Although most existing DLAs are driven by near-infrared lasers with a wavelength of approximately 1 μm, the use of long-wave infrared (LWIR) lasers at a wavelength ten times that of this wavelength indicates that it is possible to achieve excellent beam quality without sacrificing acceleration gradient. To address the lack of optimized structural designs in the LWIR band where long-distance acceleration poses unique challenges—we introduce a deep learning–based design method for LWIR dielectric grating accelerator structures. Our approach integrates geometric parameters, material properties, and optical-field energy metrics into a unified evaluation framework and uses a surrogate model to predict particle energy gain with high precision. Optimal structural parameters are then extracted to realize the final design. The simulation results show that the energy gain is 99.5 keV (a year-over-year increase of 19.9%), the transmission efficiency is 100%, the beam spot radius of 14.5 μm, and the average beam current is 20.4 fA, which is 6.9 times higher than similar near-infrared gratings, while maintaining equivalent beam brightness. This work provides a feasible technical route for designing high-netgain LWIR dielectric grating accelerators and a novel framework for optimizing the structure of complex optoelectronic devices.